Wan 2.7 on VisionX.
Alibaba’s reference-driven family: sound, an end frame, five refs — and stills.
What Wan 2.7 is on the roster
Wan 2.7 is the most directable route on the roster. It renders synchronized audio like HappyHorse 1.1, but it also takes instructions most models don’t: an end frame to land on, an audio track to drive the motion, and up to five reference images or videos for subject and style.
That makes it the one to reach for when a shot has to hit a specific mark rather than merely look right — a character from one reference moving through a scene from another, or a beat that has to end on a frame you already have.
Say where it ends
A last frame is a first-class input, so a shot can be told what to land on instead of being re-rolled until it happens to.
Five references, stills or motion
References can be video, not just images, and the prompt addresses them by number — the character from Image 1, the street from Video 1.
Audio that drives the picture
Hand it a track and the motion follows it, rather than scoring the cut afterwards and hoping the timing lands.
Your prompt, unedited
Wan offers to rewrite prompts before rendering. VisionX keeps that off, so what the composer wrote is what runs — no vendor rewrite quietly contradicting your Cast or style.
Stills from the same family
Wan 2.7 Image runs text-to-image and multi-image edits with up to nine references, and the Pro grade reaches 4K on a straight text-to-image. Same Cast, same wallet as the video route.
What Wan 2.7 actually does
The most directable route on the roster — it takes instructions most models do not accept.
Say where the shot ends
A last frame is a first-class input, so a shot can be told what to land on instead of being re-rolled until it happens to.
End frameReferences that can be motion
References can be video, not just images, and the prompt addresses them by number — the character from Image 1, the street from Video 1.
Video referencesAudio that drives the picture
Hand it a track and the motion follows it, rather than scoring the cut afterwards and hoping the timing lands.
Audio-driven motionYour prompt, unedited
Wan offers to rewrite prompts before rendering. VisionX keeps that off, so what the composer wrote is what runs — no vendor rewrite quietly contradicting your Cast or style.
No silent rewritesStills from the same family
Wan 2.7 Image runs text-to-image and multi-image edits, and the Pro grade reaches 4K on a straight text-to-image. Same Cast, same wallet as the video route.
Image routeEvery variant, exactly as the studio runs it
These rows are read live from the engine registry and the pricing engine. What you see here is what the composer quotes.
Self-serve rates run as low as $0.085 per VX at scale. The composer quotes the exact VX for your shot, references included, before you run it.
What teams route to Wan 2.7
When the shot has to hit a mark, not merely look right.
Shots that must land on a frame
A beat that has to end on an image you already have.
Character into a borrowed set
A person from one reference moving through a location from another.
Beat-matched motion
Movement timed to a track you supply rather than to a fixed cadence.
Same Cast, same wallet, different look
Route a board line to Wan 2.7, then run the identical shot on another family and compare takes. Identity lives in your Cast, not in any one model, so switching engines never strands the campaign.